Why Did Anthropic and OpenAI Both Cut AI Prices on the Same Day?

Two rival AI labs cut prices within 90 minutes of each other on Tuesday, and neither one blinked first by much. For months, Anthropic and OpenAI have competed mostly on raw capability: who can reason better, code faster, or handle a longer task without help. This week, both companies quietly admitted that price now matters just as much as power. That shift changes what businesses should compare before picking a model.

What Did Anthropic Actually Change With Opus 5.5?

Anthropic didn't just shave the price tag. Opus 5.5 costs 20% less on paper, but the real savings run deeper. Cache reads, which dominate the cost of long coding sessions, dropped 60%, from $0.50 to $0.20 per million tokens. So a typical agentic workload actually gets about 40% cheaper to run. Anthropic also removed the five-hour usage caps on its Pro, Max, and Team plans, and says the model matches its costlier Fable 5.1 tier on most everyday work. Independent safety testers, including METR, reviewed it before launch, and Anthropic says it does not cross the threshold for helping build novel weapons.

Why Did OpenAI Respond So Quickly?

Because the timing wasn't an accident. OpenAI rolled out GPT-6 Sol and GPT-6 Luna roughly 90 minutes after Anthropic's announcement went live, a gap too short to be a coincidence in an industry that watches its rivals closely. Sol now costs $2 per million input tokens, half of what its predecessor charged. Luna dropped even further, to just $0.10, aimed at high-volume, simple tasks. Both models slot in below OpenAI's flagship GPT-6 Astra, which keeps its $10/$50 price unchanged. Sam Altman announced both on social media the same afternoon. OpenAI says Sol makes roughly half as many factual errors as the model it replaces, closing in on flagship-level reliability at a fraction of the cost.

Does Cheaper Actually Mean Better?

Not automatically, and that's the part worth knowing. Independent evaluator Artificial Analysis tested GPT-6 Luna the same day and found something less flattering than OpenAI's own framing. Luna scored identically to its predecessor on general intelligence, but its coding score slipped from 43 to 41. So a lower price doesn't always come with a matching jump in capability. Anyone switching models purely to save money should test their own workload first, rather than assume "newer and cheaper" means "just as good."

How Do the New Prices Actually Compare?

Seeing all three side by side makes the shift easier to grasp than reading them one at a time:

• Claude Opus 5.5: $4 input / $20 output per million tokens, down from $5 / $25, cache reads cut to $0.20

• GPT-6 Sol: $2 input / $10 output per million tokens, down from $4 / $20

• GPT-6 Luna: $0.10 input / $0.50 output per million tokens, down from $0.20 / $1.20

• GPT-6 Astra, OpenAI's flagship, stayed unchanged at $10 / $50, the price cuts only hit the tiers below it

• All four prices above are current, standard rates, not limited-time promotions

What This Means for Businesses Running AI Workloads

If your company runs meaningful AI usage through either provider, now is the time to check your bill. A price war between the two biggest labs tends to ripple through the entire market within weeks, and every serious competitor will feel pressure to match it. Before switching models to chase savings, run your own workload against both options, since a cheaper price tag doesn't guarantee the same coding accuracy or output quality you're used to. Our generative AI services team can help you benchmark these new models against your actual use case before you commit to one.

Frequently Asked Questions

Q: Is Claude Opus 5.5 cheaper than GPT-6 Sol?

A: Per token, GPT-6 Sol is actually cheaper of the two flagship-adjacent models. But Opus 5.5 sits a tier above Sol in raw capability, so the fairer comparison is Opus 5.5 against GPT-6 Astra, where Opus 5.5 comes out well ahead on price.

Q: Which model is best for coding on a tight budget?

A: Neither Luna nor the cheapest Claude tiers are built for serious coding work. For budget-conscious coding, GPT-6 Sol or Claude Opus 5.5 are the more realistic starting points, based on their benchmark scores so far.

Q: Do these price cuts apply to ChatGPT Plus or Claude Pro subscriptions? No.

A: These are API prices, what developers and businesses pay to build on these models directly. Consumer subscription prices for ChatGPT and Claude.ai have not changed.

Q: Why did OpenAI respond within 90 minutes instead of days or weeks?

A: Both companies watch each other's announcements closely, and a same-day answer signals OpenAI didn't want Anthropic's news to dominate the news cycle alone, even briefly.